Synthetic Control Arms: An Imperfect Solution to an Impossible Question

Remember when only a decade ago, single-arm trials were all the rage?  You can go all the way back to the early 2000s to find these trial designs being used for regulatory approval.  Indeed, between 2002 and 2021, the FDA granted 176 approvals for indications within the malignant hematology and oncology area using single-arm trial designs.   About half of them (87) were for first-approval novel therapeutics, while the remainder were for supplemental indications. This is as powerful an endorsement by the FDA as one can expect.

As any industry observer would note, single-arm trial designs have increased in popularity over time. This doesn't mean that all stakeholders (regulatory, payers, physicians, and even manufacturers) don't prefer to have evidence from randomized controlled trials.  They most certainly do - and for the FDA, at least, this goes beyond a passive preference based on their guidance in late 2023.

The FDA's statement to 'unwelcome' single-arm trial evidence is a difficult acknowledgement for the industry, and none were more affected than emerging biopharmaceutical companies with their eyes set on the US market for their therapeutics addressing rare or high-mortality disease areas. 

First, let's agree that the majority of novel therapeutics approved by the FDA (and, it can be assumed for simplicity's sake, most global regulatory agencies) are originated by 'emerging biopharmaceutical' innovators.  These are the (mostly) privately-held companies that serve as pseudo-laboratories for most multinational pharmaceutical companies.

It has been many years since the discovery, pre-clinical, and early clinical study stages undertaken by major multinationals transitioned to the hotbed of biopharma innovation.  This led multinationals to shift their mode of operation from running point on discovery all the way through launch / lifecycle management to become one significantly more focused on partnership and in-licensing - followed then by launch prep and lifecycle planning.  (This is not to say that major multinationals don't discover their own successful products - they do, but just not as often as before).

This evolved model de-risked the early drug development process for multinational companies, and shifted it to privately-held emerging biopharmaceutical companies.  As for discovery, it was these small companies' entire raison d'etre.  They were, and still are, houses of 'discovery and early development', which require a 'refinancing' (capital injection and possible acquisition) to get their discoveries through clinical study and over the finish line to patients. 

Unfortunately, moving from a patent suite to enviable product in late PHASE I / early-PHASE II carries the immense risk noted above - and has become increasingly expensive to manage.  Big-brother partners (major multinationals) demand robust data - because they know the gauntlet that awaits from regulators, payers, physicians, and their own investors.  Plus, let's face it - competition is fierce.  With the advancement of platform companies over individual therapeutic products, as well as the technology that drives them, the power of choice is shifting more towards the big-brothers. 

Emerging biopharma's proverbial 'wing and prayer' commonly boiled down to one detour - the single-arm trial.  One could argue that this assortment of trial design and resulting evidence were nothing short of essential - after all, we had a limited supply of patients in rare disease categories, and there were substantial ethical concerns when operating studies for patients carrying high-mortality risk.  Regulators sensed the wave of rare / high-mortality disease therapeutics rapidly approaching, and allowing for this detour was a valid compromise between the industry and the FDA (we'll note here that they are also used and acceptable by the EMA, as well as other global regulatory agencies).  For patients, it was a gift - both post-approval, and in the trial-setting within the experimental arm of study. 

Now, let's re-calibrate to the present and face down emerging biopharma's challenge to fund studies to create the evidence required to persuade partners to help them to regulatory approval.   We should note that payers were never a fan of single-arm studies anyway (economic modelling became exceedingly challenging for them with no comparative data) - so addressing this trial design challenge will help with another key component of an eventual successful launch: payer access.  Today's answer can no longer default to simply eliminating an entire arm of study with promises of future studies and real-world evidence.

On the contrary - the solution is to buoy the quality of comparison against a defensible data set - and dare we say it, AI to the rescue. 

We will explore all of the truly stunning ways that AI and machine learning - the bleeding edge of technology - is making the process of novel therapeutic development and discovery more efficient another time.  This is not the purpose of this post. 

Instead, we will simply leave you with what appears to be next stage of evolution for trial design and execution: the growing excitement and viability around synthetic control arms.  There are a number of companies and academicians exploring this area, and if you're interested, you should study them. 

• https://www.sciencedirect.com/science/article/pii/S221133552100262X?via%3Dihub

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10785851/

• https://link.springer.com/article/10.1007/s11523-022-00910-0

• https://www.jmir.org/2023/1/e43484

• https://pmc.ncbi.nlm.nih.gov/articles/PMC10504821/

  
  

We must recognize that there are answers out there - and that pursuing such a path for an early-stage asset means leveraging real-world data collected and housed through a variety of means in less-controlled environments in order to model how patients may theoretically respond to a non-intervention in the control-arm.  AI and machine learning optimize these data, and will continue to get better at doing so as time passes.  We must get smarter about how to effectively use these techniques, and encourage stakeholder audiences to do the same. We must be the students, and then the educators.

Make no mistake, pursuing the synthetic control arm approach to trial design is a radical departure from not just randomized controlled trials, but also the alternate, nearly-conventional compromise of a single-arm trial.  Treading these synthetic waters carries risk, but there is substantial upside.  Regulatory approval, payer acceptance, and physician acknowledgement await - but just as potentially powerful is that multinational companies who are prospective development partners have greater room to operate when developing joint commercial and launch plans when they have comparative data-sets.  Beyond this, there is also a hidden benefit in the perception of partnerships with emerging biopharmaceutical companies who utilize synthetic designs: investors take note.  Wall Street has a tendency to reward companies who embrace risk and reap the rewards associated with paving new paths to commercial success. 

The sources of comparative data used in synthetic control arms are manifold, and range from your own e-health records, insurer claims data, and that watch on your wrist.  Is it perfect?  No - but that isn't what AI does - it simply enables us to do more than we may have been able to achieve on our own.  And for this particular problem, that was the highly imperfect absence of a control arm.